KR100930529B1 - Harmful video screening system and method through video identification - Google Patents

Abstract

The present invention discloses a harmful video screening system and method through video identification. The system according to the present invention includes: a harmful video index DB in which a video fingerprint of a harmful video file is indexed; A file input unit for receiving a censored video file; A video feature extracting unit extracting a video fingerprint by analyzing video data included in the input video file; And a harmful video inspection unit for searching whether a video fingerprint having a similarity with the extracted video fingerprint or more than a threshold is indexed in the harmful video index DB and inspecting whether the video file to be detected corresponds to the harmful video.

According to the present invention, since the harmful video is inspected based on the content of the video, the harmful video can be effectively blocked. In addition, if the content of the video data is the same, harmful video can be inspected even if the encoding environment of the video file is changed or the video file is secondarily edited. In addition, by combining harmful video screening and file duplication screening, efficient use of storage space is possible.

Video fingerprint, harmful video censoring, file duplication checking, video indexing

Description

System and Method for inspection of noxious moving video by video identification}

The present invention relates to a harmful video detection system, and more particularly, to a system and a method for identifying a video included in a video and detecting whether the video is harmful.

With the rapid development of information and communication technology, the speed of data transmission over the Internet is getting faster every day. Accordingly, a video streaming service for real-time education, broadcasting, movies, shopping, etc., as well as a download service of a large video file, has become possible.

The emergence of various video services has positively improved the quality of Internet services. On the other hand, illegal distribution of harmful videos using file sharing services or messenger programs linked with web hard services is emerging as another social issue. .

Currently, the most problematic video of harmful videos is pornography video. Pornography videos are legally produced, but most of them are illegally made in foreign countries or illegally shot on a video chat site or private space without the consent of the other party. In particular, in the latter case, the act of shooting a movie itself is illegal, and when a video file is distributed through the Internet, it can cause a serious mental shock to the subject, and when the subject is a celebrity, it causes a social wave. .

In recent years, digital cameras and digital camcorders that can shoot video. As camera phones are spreading, even pornographic videos in the form of User Created Contents (UCC) are being added, and the pornographic video problem becomes more serious.

It cannot be overemphasized that various harmful videos including pornography videos adversely affect the mental health of minors. In addition, when harmful videos are distributed through a public website such as an internet portal, it severely impacts the reliability and morality of the website operator. Accordingly, there is an urgent need for a technology that can effectively detect whether a video distributed through the Internet corresponds to a harmful video.

In conventional pornography video censorship technology, a video censor inspects the contents of a video manually to determine whether the video is pornographic, and analyzes the color (yellow, black, white, etc.) of the video to threshold the specific gravity of a specific color. The color screening method (Korean Patent Laid-Open Patent No. 2004-46537), which is determined as a pornographic video when it exceeds, establishes a DB for hash values of various pornographic videos, and then compares the hash value of the video to be inspected with the hash value recorded in the DB. In comparison, there is a hash value screening method (Korean Patent No. 10-517645), etc., which determines that the video is pornographic if the similarity exceeds a threshold.

However, the manual censoring method is not effective for censoring a large amount of video files because it takes a long time to perform video censorship. In addition, the color censoring method has a problem in that a censoring error occurs due to lighting of a moving picture photographing place or clothing of an actor, resulting in inferior reliability. In addition, the hash value detection method extracts some clips of pornographic video separately and edits them secondly, or censors properly when the encoding environment of the video file, for example, the type of encoding algorithm, the tag inserted in the file, the sampling rate, and the bit rate change. There is a problem that is not made.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems of the prior art, and an object thereof is to provide a reliable harmful video screening system and method based on video content of a video without depending on the encoding environment of the video file. .

Another object of the present invention is to provide a harmful video screening system and method capable of inspecting not only copied harmful video but also harmful video in various formats produced by the same video stream.

It is still another object of the present invention to provide a system and method capable of censoring secondary video edited secondarily.

Another object of the present invention is to provide a harmful video screening system and method that can reduce the burden of video screening when building a video database for providing a variety of video services.

In accordance with another aspect of the present invention, there is provided a harmful video detection system through video identification, including: a harmful video index DB in which a video fingerprint composed of a plurality of sub video fingerprints for each frame of the harmful video file is indexed; A file input unit for receiving a censored video file; A video feature extractor configured to analyze video data included in the censored video file to extract a video fingerprint composed of a plurality of sub video fingerprints per frame; And comparing the extracted video fingerprint with the fingerprint of the harmful video index DB to calculate similarity according to the degree of identity of sub-fingerprints, and if the similarity is greater than or equal to a threshold value, discriminate the censored video file as the pre-indexed harmful video. And a harmful video inspection unit which performs a censor to perform the inspection and omits indexing of the harmful video index DB of the censored video file identified as the harmful video.

If it is determined that the video file is not an indexed harmful video, the system according to the present invention displays an image included in the video file to be inspected on the screen, and the manager inspector which designates the type of the video displayed on the screen. ; And a video indexing unit which newly indexes the extracted video fingerprint into the harmful video index DB when the censored video file is designated as the harmful video through the manager inspecting unit.

According to an aspect of the present invention, a non-hazardous video index DB, the video fingerprint consisting of a sub-video fingerprint of a plurality of frames of non-hazardous video file is indexed; And if the censored video file is designated as a non-hazardous video through the manager censor, the similarity of the sub-fingerprints is calculated by comparing the extracted video fingerprint with the fingerprint of the non-harmful video index DB. When the similarity level is higher than the threshold, a censor is performed to identify the censored video file as an indexed non-hazardous video, and the non-hazardous video omitting the indexing operation on the non-harmful video index DB of the censored video file determined as non-hazardous video. The inspection unit may further include. If the non-hazardous video inspection unit determines that the censored video file is not an indexed non-hazardous video, the video indexing unit newly indexes the extracted video fingerprint in the video index DB.

According to another aspect of the present invention, if the harmful video inspection unit and the non-harmful video inspection unit, and the harmful video inspection unit discriminates the video file to be detected as analogous video, the non-hazardous video inspection request to the analogous video inspection unit The non-harmful video inspection unit may output the non-harmful video inspection according to the censoring request.

In this case, if the non-hazardous video inspection unit determines that the video object to be censored is not an indexed non-hazardous video, the manager displays an image included in the video object on the screen and specifies the type of the video displayed on the screen. Censorship; And a video indexing unit which newly indexes the video fingerprint extracted in the harmful video index DB if the censored video file is designated as a harmful video and, if the video file is designated as a non-hazardous video index, through the manager inspecting unit. It may further include;

In the present invention, the video indexer may also index the header and meta information of the video file when new indexing of the extracted video fingerprint.

Another technical problem of the present invention can be achieved by the above-described method of operating a video clip screening system through video identification, and a computer-readable recording medium which records the program.

According to the present invention, since the harmful video is inspected based on the content of the video, the harmful video can be effectively blocked. In addition, if the content of the video data is the same, harmful video can be inspected even if the encoding environment of the video file is changed or the video file is secondarily edited. Furthermore, since indexing of harmful videos is continuously performed, the burden on the censorship manager can be continuously reduced in proportion to the number of harmful videos indexed when constructing a video database for providing a video service. In addition, by combining harmful video screening and file duplication screening, efficient use of storage space is possible.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms or words used in the specification and claims should not be construed as having a conventional or dictionary meaning, and the inventors should properly explain the concept of terms in order to best explain their own invention. Based on the principle that can be defined, it should be interpreted as meaning and concept corresponding to the technical idea of the present invention. Therefore, the embodiments described in the specification and the drawings shown in the drawings are only the most preferred embodiment of the present invention and do not represent all of the technical idea of the present invention, various modifications that can be replaced at the time of the present application It should be understood that there may be equivalents and variations.

FIG. 1 is a block diagram illustrating a schematic configuration of a system for screening a harmful video through video identification according to a first embodiment of the present invention.

The harmful video screening system according to the first exemplary embodiment includes a file input unit 100, a video feature extraction unit 110, a harmful video inspection unit 120, an administrator screening unit 130, a video indexing unit 140, and a non-hazardous video. The inspection unit 150, the harmful video index DB 160, the harmful video storage 170, the non-hazardous video index DB 180, and the non-hazardous video storage 190 are included.

The file input unit 100 receives a moving image file to be censored and delivers it to the video feature extraction unit 110. Here, censorship refers to determining whether a video file corresponds to a harmful video. Harmful video refers to pornography, violence, or cruelty that has a harmful effect on adolescents. The video file has a video-only format such as mpeg, avi, mov, and the like. A video file is a file transmitted over a network. For example, the video file is a file transmitted through wired or wireless Internet, a file transmitted through mail reception, or a file transmitted through P2P communication. Alternatively, the movie file is a file stored within computer storage. In the latter case, the file input unit 100 preferably provides a user interface for the censorship manager to select at least one video file. The video file selected through the user interface is selected as a censored file. The computer on which the system according to the invention can be installed is a personal computer or a server computer. Server computers include content servers capable of wired and wireless Internet services, search portal servers, and mail servers.
For reference, the harmful video in the present invention is determined and classified on the basis of the harmfulness of the general public. For example, the computer system is designated files corresponding to the type of the harmful video by an arbitrary judgment classification of the censorship administrator, and the fingerprints of the designated files are stored in the harmful video index DB 160 to determine the harmfulness judgment of the present invention. As a reference, the fingerprint stored in the DB 160 is compared with the fingerprint of the received censored video file.

The video feature extractor 110 extracts a video fingerprint by analyzing video data included in a moving image file. Here, a video fingerprint refers to a kind of video fingerprint that can exclusively and uniquely identify video data. The video feature extractor 110 may precede the process of selectively separating only the video data from the video file before extracting the video fingerprint.

The video feature extractor 110 selectively separates a video stream from a video file to extract a video fingerprint, decodes the separated video stream, and converts the video stream into image data in a raw format. Here, the raw format means a YUV (YCbCr) format or an RGB format. However, the present invention is not limited by the type of raw format. Subsequently, the video feature extractor 110 extracts a sub video fingerprint for each frame image constituting the decoded image data.

The video feature extractor 110 extracts a sub video fingerprint for each frame image as follows. First, the video feature extractor 110 divides the frame image in block units. As an example, the frame image is divided into n × m blocks by dividing the vertical axis of the image by n and dividing the horizontal axis by m. As another example, the vertical axis of the image is divided into n blocks to divide the frame image into n blocks, and the horizontal axis is further divided into m blocks to divide the frame image into m blocks and the frame image is divided into n + m blocks.

Subsequently, the video feature extractor 110 calculates an average of Y values, which are image brightness data of each pixel, for each divided block unit, and then selects a predetermined number of bits (for example, 32 bits) based on a difference in image brightness average values between adjacent blocks. Create a sub video fingerprint of.

Equation 1 below illustrates a bit allocation function 'B (r, c, p)' that can be used to generate a sub video fingerprint for one frame image. For reference, Equation 1 below assumes a case in which one frame image is divided into n × m blocks.

Here, B (r, c, p) is a function for allocating bits to blocks located in the r-th and c-th based on the horizontal and vertical axes of the p-th frame image. F (r, c, p) represents an average of image brightness values of blocks located at r-th and c-th, respectively, based on the horizontal and vertical axes of the p-th frame image.

In F (r, c, p) of Equation 1, r has an integer value from 1 to n, and c has an integer value from 1 to m-1. Therefore, when the frame image is divided into 4 × 9 = 36 blocks, the sub video fingerprint has a code string of 4 × 8 = 32 bits. Α is a constant smaller than 1 and can be appropriately selected. Α is 0.95, for example.

The video feature extractor 110 may generate a sub video fingerprint by using a Cb or Cr value, which is color difference data of each pixel, in addition to the Y value, which is brightness data of an image. It will be apparent to those skilled in the art that a method for generating a sub video fingerprint using a Cb or Cr value will be substantially the same as described above.

After extracting the sub video fingerprint for each frame image constituting the video, the video feature extractor 110 completes the generation of the video fingerprint for the video file by grouping each extracted sub video fingerprint.

The above-described video fingerprint extraction technique is disclosed in Job Oostveen, Ton Kalker, Jaap Haitsma's article "Feature Extraction and a Database Strategy for Video Fingerprinting" (Lecture Notes in Computer Science 2314 Springer 2002, pp. 117-128). Accordingly, all content of the article can be incorporated as part of the construction of the present invention. On the other hand, the present invention is not limited by the specific method of extracting the video fingerprint, any algorithm that can extract the video fingerprint can be employed.

The harmful video inspection unit 120 receives a video fingerprint extracted from the video feature extraction unit 110. Then, by checking whether the video fingerprint corresponding to the extracted video fingerprint is indexed in the harmful video index DB 160, whether the video file corresponds to the harmful video is detected.

If there is a video fingerprint corresponding to the extracted video fingerprint, the video file corresponds to the harmful video. On the contrary, if there is no video fingerprint corresponding to the extracted video fingerprint, the video file is at least indexed to the harmful video. Not applicable The case where the video fingerprints correspond to each other means that the similarity between the video fingerprints exceeds a threshold. Accordingly, when the video file is determined to be a harmful video, it means that a video fingerprint whose similarity with the extracted video fingerprint exceeds a threshold is already indexed in the harmful video index DB 160.

When the harmful video inspection unit 120 calculates the similarity between the video fingerprint extracted by the video feature extraction unit 110 and the video fingerprint indexed in the harmful video index DB 160, Is the same as

First, the harmful video inspection unit 120 designates a video fingerprint extracted by the video feature extraction unit 110 as a query fingerprint, and uses the video fingerprint indexed in the harmful video index DB 160 as the comparison target fingerprint. Specify. Thereafter, the sub video fingerprints included in the query fingerprint are sequentially extracted in the frame order to determine whether the same sub video fingerprint as the extracted sub video fingerprint exists in the comparison target fingerprint. As a result, when it is determined that the same sub video fingerprint exists, the harmful video inspection unit 120 determines a predetermined number (eg, 100) before and after a frame where the sub video fingerprint matches from the query fingerprint and the fingerprint to be compared. Extract a plurality of sub video fingerprints for the < RTI ID = 0.0 > Then, the similarity between the query fingerprint and the fingerprint to be compared is determined by comparing the extracted sub video fingerprint groups with each other.

For example, the harmful video inspection unit 120 compares the sub video fingerprint groups extracted from the query fingerprint and the comparison target finger, calculates a ratio of matching sub video fingerprint pairs, and sets the value in advance. The similarity between the query fingerprint and the fingerprint to be compared is determined by comparing with a threshold.

As another example, the file duplication checker 30 may determine the overall average bit error rate based on a bit error rate between the sub fingerprints corresponding to the sub video fingerprint groups extracted from the query fingerprint and the comparison target fingerprint. After calculating, compare the size with a preset threshold to determine the similarity between the query fingerprint and the fingerprint to be compared.

Meanwhile, in addition to the above-described method, the harmful video inspection unit 120 simply calculates a ratio of the sub video fingerprints included in the comparison target fingerprint among the sub video fingerprints included in the query fingerprint, and then sets a preset threshold value. The degree of similarity between the query fingerprint and the comparison target fingerprint may be determined in comparison with.

As described above, by determining whether a moving image file is a harmful video using a video fingerprint, effective harmful video censoring is possible. If the video fingerprint extracted from the video file has a similarity above the threshold with the video fingerprint already indexed, the video content of the video file is the same as the video content of the video file that is already identified as the harmful video, and the video content is the same. This is because censored video files can be regarded as harmful videos.

The harmful video inspection unit 120 skips the indexing of the video file when the video file input through the file input unit 100 is identified as the pre-indexed harmful video. Instead, the extracted video fingerprint and the video file may be stored in separate storage spaces of the harmful video index DB 160 and the harmful video storage 170, respectively. It is apparent that the separate storage space may be set in a separate DB and storage from the harmful video index DB 160 and the harmful video storage 170.

When the harmful video inspection unit 120 determines that the video file is not a pre-indexed harmful video as a result of the video file based on the video fingerprint, the harmful video inspection unit 120 performs manual inspection by the censorship manager 130. You can optionally proceed with the process.

That is, the manager censor 130 receives a manual censor for the video file from the harmful video inspecting unit 120. Then, the manager censor 130 visually displays the video file determined to be not the indexed harmful video to the censor and receives the type of the video from the censor.

The visual display of the moving image file may be performed by playing the moving image file or capturing and continuously arranging still images of the moving image at regular frame intervals from the moving image file. The latter display method has the advantage that the video censoring time required is small because the entire contents of the video can be checked at a glance. However, the present invention is not limited to the visual display method of the video file.

The inspector inspecting unit 130 may perform the following process according to how the inspector designates the type of the video file. That is, when the censorship manager designates a video file as a harmful video, the manager censor 130 requests a new indexing of the video file to the video indexer 140. Then, the video index unit 140 newly indexes the video fingerprint extracted from the video file to the harmful video index DB 160. Furthermore, the video file indexed in the harmful video storage 170 is stored. Meanwhile, although not essential, the video indexer 140 may also index the header and meta information extracted from the video file when the video fingerprint is indexed.

Conversely, if the censorship manager designates a video file as a video, the manager censor 130 requests a new indexing as a video by analogy to the video indexer 140. Then, the video indexing unit 140 newly indexes the video index DB 180 using the video fingerprint extracted from the video file. Furthermore, the video file indexed in the analog video storage 190 is stored. Meanwhile, although not essential, the video indexer 140 may also index the header and meta information extracted from the video file when the video fingerprint is indexed.

Unlike the above description, when the video file is designated as a video, the manager inspecting unit 130 may perform a file redundancy screening process prior to DB indexing and storage of the video file. That is, if the inspector inspector 130 designates a video file as a video clip, the manager inspector 130 requests a duplicate file inspection to the video clip inspector 150. Then, the non-hazardous video inspection unit 150 searches whether the video fingerprint corresponding to the video fingerprint extracted from the video file is indexed in the non-hazardous video index DB 180 and performs file duplication inspection. Here, the concept of correspondence between video fingerprints has been described above, and file duplication detection may be implemented by the same principle as harmful video detection.

As a result of the file redundancy inspection, if it is determined that the video fingerprint whose similarity with the extracted video fingerprint is greater than or equal to the threshold is not indexed in the analogous video index DB 180, the non-hazardous video inspection unit 150 determines whether the video fingerprint unit ( 140) request new indexing of the video file. Accordingly, the moving picture indexer 140 newly indexes the extracted video fingerprint in the moving picture index DB 180 and stores the moving picture file in the moving picture storage 190.

On the contrary, if it is determined that the video fingerprint having a similarity with the extracted video fingerprint is greater than or equal to the threshold, the video fingerprint DB 150 determines that the video file and the video contents are substantially indexed. It is assumed that the duplicate video file is similarly stored in the video storage 190. In this case, the DB indexing of the video fingerprint and the storage of analogous video storage are omitted. Instead, the extracted video fingerprint and the video file may be stored in separate storage spaces of the non-hazardous video index DB 180 and the non-hazardous video storage 190, respectively. It is apparent that the separate storage space may be set in a separate DB and storage from the non-hazardous video index DB 180 and the non-hazardous video storage 190.

2 is a block diagram illustrating a schematic configuration of a system for screening a harmful video through video identification according to a second embodiment of the present invention.

The harmful video screening system according to the second embodiment includes the same configuration as the system according to the first embodiment described above, and includes the harmful video screening unit 120, the manager screening unit 130, and the non-hazardous video screening unit 150. The order is different.

That is, in the first embodiment, the components are arranged in the order of the harmful video inspection unit 120, the manager inspecting unit 130, and the non-hazardous video inspection unit 150, but in the second embodiment, the harmful video inspection unit 120, analogy. The components are arranged in the order of the video inspection unit 150 and the manager inspector 130. Therefore, the harmful video screening system according to the second embodiment continuously performs harmful video screening and file duplication screening, and finally performs manual screening by the screening manager. On the other hand, the configuration itself of performing harmful video screening and file duplication screening by calculating the similarity between video fingerprints is found to be substantially the same as the first embodiment described above.

Specifically, when the harmful video inspection unit 120 determines that there is no video fingerprint corresponding to the video fingerprint extracted from the video file as a result of searching the harmful video index DB 160, the harmful video inspection unit 150 compares the harmful video inspection unit 150. ), We ask for censoring duplicate files based on the video.

Then, the non-hazardous video inspecting unit 150 may generate a video fingerprint (ie, a comparison target fingerprint) of each video file stored in the non-hazardous video index DB 180 and a video fingerprint extracted from the video feature extracting unit 110 (ie, , The query fingerprint) is compared with each other, and whether the comparison fingerprint corresponding to the query fingerprint is compared is searched in the video index DB 180 to perform file duplication inspection.

As a result of file duplication inspection, if it is determined that the video fingerprint corresponding to the extracted video fingerprint is already indexed in the analogous video index DB 180, the analogous video inspection unit 150 determines the contents of the video file and the video. It is assumed that this substantially duplicate video file is similarly stored in the video storage 190. In this case, the DB indexing of the video fingerprint and the storage storage of the video file are omitted. Instead, the extracted video fingerprint and the video file may be stored in separate storage spaces of the non-hazardous video index DB 180 and the non-hazardous video storage 190, respectively. It is apparent that the separate storage space may be set in a separate DB and storage from the nonhazardous video index DB 180 and the nonhazardous video storage 190.

On the other hand, if it is determined that the video fingerprint corresponding to the extracted video fingerprint is not indexed in the analogous video index DB 180, the non-hazardous video inspecting unit 150 is the inspector manager on the manager censor 130 side. Request manual censorship by Then, the manager censor 130 visually displays the video file to the censorship manager and receives a video type from the censorship manager. Here, the visual display method of the video file has been described above.

If the inspector designates the video file as a harmful video, the inspector inspecting unit 130 requests the video indexing unit 140 to newly index the video file. Then, the video index unit 140 newly indexes the video fingerprint extracted from the video file to the harmful video index DB 160. Furthermore, the video file indexed in the harmful video storage 170 is stored. Meanwhile, although not essential, the video indexer 140 may also index the header and meta information extracted from the video file when the video fingerprint is indexed.

On the contrary, when the censorship manager designates a video file as a video clip, the manager censor 130 requests a new indexing as a video clip on the video indexing section 140 side. Then, the video indexing unit 140 newly indexes the video index DB 180 using the video fingerprint extracted from the video file. Furthermore, the video file indexed in the analog video storage 190 is stored. Meanwhile, although not essential, the video indexer 140 may also index the header and meta information extracted from the video file when the video fingerprint is indexed.

The harmful video screening system according to the second embodiment of the present invention performs the manual video screening by the censorship manager after continuously performing the harmful video screening and the file duplication screening, so that the burden of the censorship manager can be reduced compared to the first embodiment. There is this.

3 and 4 are flowcharts illustrating a method of operating a harmful video detection system through video identification according to a first embodiment of the present invention. The execution subject of each step is the harmful video screening system shown in FIG.

First, in step S101, a harmful video index DB in which video fingerprints of a plurality of harmful video files are indexed, and a non-hazardous video index DB in which video fingerprints of a plurality of non-hazardous video files are indexed are provided.

In step S102, a video file to be inspected is input while a harmful video index DB is provided. Here, the video file is input through a network or from storage inside the computer.

In step S103, the video stream is separated from the input video file. Then, in step S104, the separated video stream is analyzed to extract a video fingerprint. The extraction method of the video fingerprint has already been described above.

In step S105, the harmful video index DB is queried to determine whether a video fingerprint having a similarity to the extracted video fingerprint or more than a threshold is indexed in the harmful video index DB.

Step S106 is a step that proceeds when the video fingerprint having a similarity or higher than a threshold with the video fingerprint extracted in the harmful video index DB is determined as a result of the determination in step S105, and the video file is converted into the harmful video. The new indexing of the extracted and extracted video fingerprint and the storage storage of the video file are omitted.

In contrast, step S107 is a step that is performed when a video fingerprint having a similarity or higher than a threshold with the video fingerprint extracted in the harmful video index DB is not indexed as a result of the determination in step S105, The included video is displayed to the censorship manager. Here, the display of the image may be performed by a method of playing a moving image file or a method of capturing and continuously arranging still images of a moving image at regular frame intervals.

In step S108, the censor inspects the visually displayed image and then determines the type of the video.

Step S109 is a step that is performed when the type of the video is the harmful video as a result of the determination in step S108. The extracted video fingerprint is newly indexed into the harmful video index DB, and the video file is newly added to the storage. Save it. Here, when newly indexing the video fingerprint, the header and meta information of the video file may be indexed together.

On the contrary, step S110 is a step that is performed when the type of the video is not a harmful video as a result of the determination in step S108, and has similarity between the video fingerprint extracted by querying the analogous video index DB and having a similarity or more than a threshold value. It is determined whether the video fingerprint is indexed in the analogous video index DB, and duplicate file inspection is performed.

Step S111 is a step that proceeds when the duplicate fingerprint detection result in step S110 is similar to the video fingerprint extracted in the video index DB and the video fingerprint having a similarity or higher than a threshold is indexed. The new indexing of the extracted video fingerprint and the storage of the video file are omitted. Then terminate the duplicate file screening process. Alternatively, the extracted video fingerprint and the video file may be moved to a separate storage space allocated to the video index DB and the non-harmful video storage, and then the duplicate file screening process may be terminated.

On the contrary, step S112 is a step that proceeds when the result of duplicate file inspection in step S110 is that a video fingerprint having a similarity or higher than a threshold with the video fingerprint extracted in the analogous video index DB is not indexed. The video file is identified as a new analogous video, and the extracted video fingerprint is likened to a new video index DB, and the video file is similarly stored in the video storage. Here, when newly indexing the video fingerprint, the header and meta information of the video file may be indexed together.

As described above, when the process of duplication of file screening based on harmful video censoring, manual censoring by a censoring manager, and a non-hazardous video is completed, the entire process of the harmful video censoring method according to the first embodiment of the present invention is completed.

5 and 6 are flowcharts illustrating a method of operating a harmful video detection system through video identification according to a second embodiment of the present invention. The execution subject of each step is the harmful video screening system shown in FIG.

As shown in the figure, a step of providing a video index DB similar to the harmful video index DB (step S101), a step of receiving a censored video file (step S102), and a step of separating a video stream from the video file (S103). Step), extracting the video fingerprint from the separated video stream (step S104), censoring the video using the harmful video index DB (step S105), and extracted video when the video file is determined to be harmful video The step of omitting the new indexing of the fingerprint and the storage storage of the video file (step S106) is substantially the same as the harmful video screening method according to the first embodiment described above. However, if the video file is not determined to be a harmful video, the subsequent process is different from that of the first embodiment.

That is, in step S105, if the moving image file is not discriminated as harmful video, the process proceeds to step S113. In step S113, the non-hazardous video index DB is searched to determine whether a video fingerprint having a similarity or higher than the extracted video fingerprint is indexed in the non-hazardous video index DB, and duplicate file inspection is performed. .

Step S114 is a step that proceeds when the duplicate fingerprint detection result in step S113 is similar to the video fingerprint extracted in the video index DB and the video fingerprint having a similarity or higher than a threshold is indexed. The new indexing of the extracted video fingerprint and the storage of the video file are omitted. Alternatively, the extracted video fingerprint and video file can be moved to a separate storage space allocated to the video index DB and analog video storage.

On the contrary, step S115 is a step that proceeds when the duplicate fingerprint detection result in step S113 does not index the video fingerprint extracted in the non-harmful video index DB and the video fingerprint having a similarity or higher than a threshold is indexed. The video contained in the video file is displayed to the censorship manager. Here, the display of the image may be performed by a method of playing a moving image file or a method of capturing and continuously arranging still images of a moving image at regular frame intervals.

In step S116, the censor inspects the visually displayed image and then determines the type of the video.

Step S117 is a step that proceeds when the type of the video is the harmful video as a result of the determination in step S116. The extracted video fingerprint is newly indexed into the harmful video index DB, and the video file is stored in the harmful video storage. Save new.

On the contrary, step S118 is a step that proceeds when the video type is analogous as a result of the determination in step S116. The extracted video fingerprint is newly indexed in the video index DB and the video file is compared. Similarly store in new video storage. Here, when newly indexing the video fingerprint, the header and meta information of the video file may be indexed together.

As described above, when the harmful video censoring, the duplicate video screening based on the non-hazardous video, and the manual censoring process by the censorship manager are finished, the entire process of the harmful video censoring method according to the second embodiment of the present invention is completed.

The operating method of the harmful video screening system according to the present invention may be implemented in the form of program instructions that can be executed by various computer means and recorded in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the present invention, or they may be of the kind well-known and available to those skilled in the computer program arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Hardware devices specially configured to store and execute program instructions such as magneto-optical media and ROM, RAM, flash memory and the like. The medium may be a transmission medium such as an optical or metal wire, a waveguide, or the like including a carrier wave for transmitting a signal specifying a program command, a data structure, or the like. Examples of program instructions include machine code, such as produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

7 is an internal block diagram of a general-purpose computer system that can be employed to perform the method of operation of the harmful video screening system according to the present invention.

The general purpose computer system 400 includes one or more processors 410 connected to a main memory including a RAM 420 and a ROM 430. The processor 410 is also called a central processing unit (CPU). As is well known in the art, the ROM 430 serves to transmit data and instructions to the processor 410 unidirectionally, and the RAM 420 typically transfers data and instructions to the processor 410 bidirectionally. Used to transmit RAM 420 and ROM 430 may include any suitable form of computer readable media. The mass storage device 440 is bidirectionally coupled to the processor 410 to provide additional data storage capability, and may be any of the computer readable recording media described above. The mass storage device 440 is used to store programs, data, and the like, and is a secondary memory device such as a hard disk which is generally slower than the main memory device. Certain mass storage devices such as CD ROM 460 may be used. The processor 410 may include one or more input / output interfaces, such as video monitors, trackball mice, keyboards, microphones, touchscreen displays, card readers, magnetic or paper tape readers, voice or handwriting readers, joysticks, or other known computer input / output devices. 450). Finally, the processor 410 may be connected to a wired or wireless communication network through the network interface 470. Through this network connection, the procedure of the method described above can be performed. The apparatus and tools described above are well known to those skilled in the computer hardware and software arts. On the other hand, the hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention.

As described above, although the present invention has been described by way of limited embodiments and drawings, the present invention is not limited thereto and is intended by those skilled in the art to which the present invention pertains. Of course, various modifications and variations are possible within the scope of equivalents of the claims to be described.

The following drawings, which are attached to this specification, illustrate exemplary embodiments of the present invention, and together with the detailed description of the present invention serve to further understand the technical spirit of the present invention, the present invention includes matters described in such drawings. It should not be construed as limited to.

1 is a block diagram showing a schematic configuration of a system for screening a harmful video through video identification according to a first embodiment of the present invention.

2 is a block diagram illustrating a schematic configuration of a system for screening a harmful video through video identification according to a second embodiment of the present invention.

3 and 4 are flowcharts illustrating a method of operating a harmful video detection system through video identification according to a first embodiment of the present invention.

5 and 6 are flowcharts illustrating a method of operating a harmful video detection system through video identification according to a second embodiment of the present invention.

7 is an internal block diagram of a general-purpose computer system that can be employed to perform the method of operation of the harmful video screening system according to the present invention.

<Main reference number in drawing>

100: file input unit 110: video feature extraction unit

120: harmful video inspection unit 130: administrator inspector

140: video index unit 150: analogous video inspection unit

160: harmful video index DB 170: harmful video storage

180: non-hazardous video index DB 190: non-hazardous video storage

Claims (16)

A harmful video index DB in which a video fingerprint including a plurality of sub video fingerprints for each frame of the harmful video file is indexed; A file input unit for receiving a censored video file; A video feature extractor configured to analyze video data included in the censored video file to extract a video fingerprint composed of a plurality of sub video fingerprints per frame; And Comparing the extracted video fingerprint and the fingerprint of the harmful video index DB to calculate the similarity according to the degree of identity of the sub-fingerprints, and if the similarity is equal to or greater than the threshold value to determine the censored video file as pre-indexed harmful video And a harmful video inspection unit that performs the inspection and skips the indexing operation on the harmful video index DB of the censored video file determined as the harmful video. delete The method of claim 1, If the harmful video inspection unit determines that the video file to be inspected is not indexed harmful video, An administrator inspecting unit which displays an image included in a censored video file on a screen and receives a type of the video displayed on the screen; And And a video indexing unit for newly indexing the extracted video fingerprint into a harmful video index DB when the video object to be inspected is designated as a harmful video through the inspector inspecting unit. Screening system. The method of claim 3, A non-hazardous video index DB in which a video fingerprint composed of a plurality of frame-specific sub video fingerprints of the non-hazardous video file is indexed; And If the censored video file is designated as a non-hazardous video through the inspector inspecting unit, the similarity according to the degree of identity of the sub-fingerprints is calculated by comparing the extracted video fingerprint with the fingerprint of the non-hazardous video index DB. When the similarity is greater than or equal to the threshold value, a screening process is performed to discriminate a censored video file as an indexed non-hazardous video, and the indexing operation for omitting indexing of a non-harmful video index DB of the censored video file determined as a non-hazardous video is omitted. Harmful video screening system through video identification, characterized in that it further comprises a video inspection unit. The method of claim 4, wherein If the non-hazardous video inspecting unit determines that the video file to be inspected is not an indexed non-hazardous video, the video indexing unit newly indexes the extracted video fingerprint in the video index DB, and thus the harmful video through the video identification. Video censorship system. The method of claim 1, A non-hazardous video index DB in which a video fingerprint composed of a plurality of frame-specific sub video fingerprints of the non-hazardous video file is indexed; And By comparing the extracted video fingerprint with the fingerprint of the non-hazardous video index DB, a similarity is calculated according to the degree of identity of sub-fingerprints. And a non-hazardous video inspection unit which performs a discriminating inspection and omits indexing of the non-hazardous video index DB of the censored video file determined as non-hazardous video. If the harmful video inspection unit discriminates the video object to be censored as an analogous video, outputs a video to the analogous video inspection request to the analogous video inspection unit, The harmful video inspection unit harmful video detection system through video identification, characterized in that for performing the harmful video inspection according to the censoring request. delete The method of claim 6, If the non-harmful video inspection unit determines that the video file to be inspected is not an indexed non-harmful video, An administrator inspecting unit which displays an image included in a censored video file on a screen and receives a type of the video displayed on the screen; And A video indexing unit for newly indexing a video fingerprint extracted in a harmful video index DB if the censored video file is designated as a harmful video and a video fingerprint extracted in a non-hazardous video index DB if it is designated as a non-hazardous video; Harmful video screening system through video identification, characterized in that it further comprises. (a) providing a harmful video index DB indexed with a video fingerprint composed of a plurality of frame-specific sub video fingerprints of the harmful video file; (b) receiving a censored video file; (c) extracting a video fingerprint including a plurality of sub video fingerprints per frame by analyzing video data included in the censored video file; And (d) comparing the extracted video fingerprint with the fingerprint of the harmful video index DB to calculate similarity according to the degree of identity of sub-fingerprints, and if the similarity is equal to or greater than a threshold value, the harmful video pre-indexed for the censored video file. And performing an indexing operation for determining whether the video is detected and skipping indexing of the harmful video index DB of the video object to be inspected determined as the harmful video. delete The method of claim 9, In step (d), if it is determined that the censored video file is not an indexed harmful video, Displaying an image included in a censored video file on a screen; Receiving a designation of the type of the video displayed on the screen; And And if the type of the designated video is a harmful video, newly extracting the extracted video fingerprint into a harmful video index DB. The method of claim 11, Providing a non-hazardous video index DB in which a video fingerprint composed of a plurality of frame-specific sub video fingerprints of the non-hazardous video file is indexed; And When the censored video file is designated as a non-hazardous video, the extracted video fingerprint and the fingerprint of the non-hazardous video index DB are compared to calculate similarity according to the degree of equality of sub-fingerprints, and the similarity is greater than or equal to a threshold. And performing a censor to determine the censored video file as an indexed non-hazardous video and omitting the indexing operation on the non-hazardous video index DB of the censored video file determined as non-hazardous video. A harmful video detection method through video identification. The method of claim 12, As a result of the non-harmful video censoring, if it is determined that the censored video file is not an indexed non-harmful video, harmful video censoring through video identification is performed by newly indexing the extracted video fingerprint into a video index DB. Way. The method of claim 9, Providing a non-hazardous video index DB in which a video fingerprint composed of a plurality of frame-specific sub video fingerprints of the non-hazardous video file is indexed; In step (d), if the censored video file is identified as an analogous video, the extracted video fingerprint and the fingerprint of the analogous video index DB are compared to calculate similarity according to the degree of identity of the sub-fingerprints. Performing censoring to identify the censored video file as an indexed non-hazardous video when the similarity is greater than or equal to a threshold, and skipping indexing of the non-harmful video index DB of the censored video file determined as non-hazardous video. ; The harmful video detection method through video identification, characterized in that it further comprises. delete The method of claim 14, Displaying the image included in the censored video file on the screen when it is determined that the censored video file is not an indexed non-harmful video as a result of the non-harmful video inspection; Receiving a designation of the type of the video displayed on the screen; And And further indexing the video fingerprint extracted in the harmful video index DB when the censored video file is designated as a harmful video and in the non-hazardous video index DB when it is designated as a non-hazardous video. How to detect harmful videos by video identification.

Images